Sentiment Analysis

When you start looking into sentiment analysis, you’re likely to come across a lot of experts who tell you it’s more than determining positive, negative or neutral opinions. Then they’ll launch into a discussion of how they determine positive, negative, and neutral opinions.

Luckily, some go a step further and discuss graded sentiment analysis which explores the strength of the positive or negative sentiment. They may go even further and explain which positive (e.g., “happy,” ”relieved”) or negative (“angry,” “betrayed”) emotions are expressed. Some go beyond assessment of the overall sentiment to tell you which aspects of the subject generate positive or negative opinions, and some can tease out intentions, e.g., whether the positive sentiments add up to a likely future purchase. The more deeply you understand customer opinions, the more you can do with them.

Even if one can manage to gather all the potential sources of sentiment – online reviews, social media, customer satisfaction surveys, and so on – just reading through it all is impossible for a human. Add in translation (if multi-lingual), interpreting sarcasm and jargon, summarizing findings, and convincing others of your recommendations escalates the process from merely impossible to....super-impossible.

However, you probably already guessed technology can execute sentiment analysis “at scale,” but it can also provide more consistent analysis. If you assign multiple humans to analyze text (or images) for emotional content, a 20-35% disagreement (depending on whom you ask) is likely to occur. Using technology eliminates this problem because every interpretation will apply the same lessons in the same way.

Of course, the big question is whether these interpretations are consistently better or consistently worse, and this will depend on how well the sentiment analysis solution comprehends natural language (regardless of which language is used). The same condition applies to human analysts. They may have a head start in how to parse a sentence into clauses, the meaning of slang (and rhyming slang) terms, negative words used in a positive way, and so on, but machines can be taught, and they are learning.

Customer (and employee, etc.) opinions are everywhere whether you ask for them or not, and they present opportunities to those who know how to find them and risks to those who ignore them. They can be found in public places, such as product reviews on Walmart.com or Amazon.com, as well as your proprietary sources, such as customer support logs or employee feedback. Sentiment analysis not only “takes the temperature” of different populations but gives you the information on whether to make it hotter or colder and how.

Brand managers can learn how customers (and non-customers) interpret different messages or react to events. Product managers can learn which features are most valued and what unmet needs exist. Ecommerce managers can learn which messages and features to amplify to improve business results.

The applications and implications of sentiment analysis are as limitless as the feedback that is constantly being produced and shared.